38 replies to this topic

[Xerxes]

Hello Sergei,

some additional results:

FX-4500PA: 159 sec
TI-95: 114 sec
HP-28S: 29.5sec

[Xerxes]

P1:

AC

1.0000001 Kin1 Kin2

27 Min



P2:

Kout1 * Kout1 = Kin1

Kout2 x^y 2.01 = Kin2

1 M- MR x>0

[DrCoyote]

I -need- an MK-152!

[verena]

Casio ?x-3600P

P1:
KAC Min

P2:
Kout1  Kin*1
Kout2  x[sup]y[/sup]  2.01  =  Kin2
1  M-  x<=M
1.0000001 Kin1 Kin2
27 Min 
1 Kin-3 Kout3 x>0
10 Kin3
ENT
x>0

Useage: Run P1, then P2. When "10 _{ENT P2}" appears, "RUN" and start timing, or enter zero to leave the program.

Casio fx-3600P - 217 seconds

Edited by verena, 19 March 2008 - 12:51 PM.

[Xerxes]

Nice solution. On the FX-3900PV it takes 56.0 seconds.

[verena]

Nice solution. On the FX-3900PV it takes 56.0 seconds.

Thank you, Xerxes.
Сергей, please remove the two stars ** and the comment about "this Casio only capable of one loop" from your web page, thank you.
And here are the results of my slowest programmable calculator:
Casio fx-201P

0 = K10: ST# 1: 1 = K1.0000001: 2 = 1: 3 = K0: 
ST# 2: 1 = 1 * 1: 2 = 2 x^y K2.01: 3 = 3 + K1: IF 3 = K27: 2: 3: 3:
ST# 3: 0 = 0 - K1: IF 0 = K0: 4: 4: 1: ST# 4: ANS 1: 2:

It takes 10 minutes and 31 seconds = 631 seconds before displaying ANS 1 = 674432.8206 and ANS 2 = 3070869.6

[Xerxes]

A better two stars comment would be: "this Casio only capable of one label"

Verena, are you interested in a more challenging problem for the FX-201P?

[verena]

I think "cycle" means "loop", not "label". The fx-3600P doesn't have labels. Anyway, the entire program (double loop) fits in an fx-3600P, that's why I requested the correction from Sergei.

Re: a "more challenging problem" - you know, nothing is a problem for the powerful Casio fx-201 with its 127 steps of program memory and extensive portfolio of transcendental, numerical and logical functions, as long as it doesn't have to work for more than 8 hours, the batteries will be empty and I don't have an AC adapter. Bring it on !

[Xerxes]

I think "cycle" means "loop", not "label". The fx-3600P doesn't have labels. Anyway, the entire program (double loop) fits in an fx-3600P, that's why I requested the correction from Sergei.

Yes, of course you are right with the lack of labels. What I meant was that there
is only one "pseudo label" at the beginning of each programm area that makes it a
bit tricky to have nested loops. Sorry for the unclear statement.

Re: a "more challenging problem" - you know, nothing is a problem for the powerful Casio fx-201 with its 127 steps of program memory and extensive portfolio of transcendental, numerical and logical functions, as long as it doesn't have to work for more than 8 hours, the batteries will be empty and I don't have an AC adapter. Bring it on !

I have used the n-queens problem to test the speed of programmable calculators
and pocket computers. It's a test for the basic programming elements and not a
mathematical test. It's an iterative backtracking algorithm.

This is the version for the FX-4500P:

Mcl
 Fixm
 R=8  (8 queens)
 Lbl 0
 X=R=>Goto 4_
 X=X+1
 A[X]=R
 Lbl 1
 S=S+1
 Y=X
 Lbl 2
 Y=Y-1
 Y=0=>Goto 0_
 T=A[X]-A[Y]
 T=0=>Goto 3_
 X-Y<>Abs T=>Goto 2_
 Lbl 3
 A[X]=A[X]-1
 Ans<>0=>Goto 1_
 X=X-1
 X<>0=>Goto 3_
 Lbl 4
 S

And this is my first try to translate the algorithm to the
unusual language of the FX-201P:

MAC
ST#0: IF 0=K6:5:4:4:  (6 queens)
ST#5: 0=0+K1:
	  I=0:
	  IM=K6:  (6 queens)
ST#1: 8=8+K1:
	  9=0:
ST#2: 9=9-K1:
	  IF 9=K0:0:0:6:
ST#6: I=0:
	  7=IM:
	  I=9:
	  7=7-IM:
	  IF 7=K0:7:3:8:
ST#7: 7=7+/-:
ST#8: 7=7+9-0:
	  IF 7=K0:2:3:2
ST#3: I=0:
	  IM=IM-K1:
	  IF IM=K0:1:9:1:
ST#9: 0=0-K1:
	  IF 0=K0:3:4:3:
ST#4: Ans8:

This version is for 6 queens because the FX-201P dosn't have enough variables.
If the program works correctly the output has to be 171 nodes.
For testing the correctness of the program on very slow calculators, I use 4 queens
with 26 nodes.

Does the program fit in to the 127 steps? If no, have you any suggestions to
make it possible? If it works, can you please measure the execution speed for 6 queens?
I guess it will take about 20-40 minutes.

Please let me know, if you need more information or have a look at the link
in my signature. Thank you.

[verena]

Does the program fit in to the 127 steps?

No, too big, I stopped counting at 140 at about 6 to 8 lines before the end. Each IF with a K takes at least 12 steps, labels take 3 steps...

If no, have you any suggestions to make it possible? If it works, can you please measure the execution speed for 6 queens?

I'll try to figure out how it works, program it if possible, or not, and let you know. Could be a couple weeks.

Edit - I know the rules for chess, but what are the "nodes" you are talking about ? "S" in the fx-4500P listing ? And, I'm getting quite pessimistic about getting your n-queens problem in the fx-201P...

Sergei, in your table please correct the times for the fx-3600P = 217 seconds, and fx-201P = 631 seconds, thank you.

[Xerxes]

No, too big, I stopped counting at 140 at about 6 to 8 lines before the end. Each IF with a K takes at least 12 steps, labels take 3 steps...

I cannot understand why Casio used such a wasteful language in a calculator with only 127 steps.
The same program on a FX-502P/602P/603P needs only 48 steps and btw 143 on the FX-4500P.

I know the rules for chess, but what are the "nodes" you are talking about ? "S" in the fx-4500P listing ?

The nodes in the backtracking search tree.

And, I'm getting quite pessimistic about getting your n-queens problem in the fx-201P...

I guess it's almost impossible unfortunately. Can you please run this test instead?

MAC
ST#0:
0=0+K1
IF 0=K0:0:0:0

What is the content of M0, if you interrupt the execution after 60 seconds?

Thank you.

[verena]

Xerxes,
Good news, I managed to get the 6-queen problem in 126 steps and the fx-201P takes 7 minutes and 38 seconds to solve it.
There was no space keft for a node counter but I'm sure it did 171 as I parallel developed the algorithm on an fx-7000G.
And the results are right too, ANS1-ANS6 being 5, 3, 1, 6, 4, 2.

Edit: It is possible to do your web page's 8-queen problem as well on the fx-201P, I'll keep you updated. runtime is around 54 minutes.

Note on the side, how easy the fx-7000G is to program compared to the modern graph Casios - all commands right at your fingertips, no OPTN menus to wade through, etc.

Edited by verena, 03 April 2008 - 04:27 AM.

[Xerxes]

I'm really impressed! Thanks for your effort.

Yes, in fact the early CASIOs with formula programming are more practical
in some cases.

[DrCoyote]

I had a difficult time measuring the results on my TI-Nspire CAS. Like the FX-9860G, the time is very short. After averaging several runs, I get approximately 2.1 seconds.

To facilitate measuring run time a bit better, I added another loop to run the whole algorithm 10 times. With that set-up, I got run times of about 18.5 seconds. At first glance, this might seem like a speed of 1.85 seconds, but that did not, obviously, include time for whatever housekeeping goes on before running a program.

Oh, and I have a Durabrand 828 graphing calculator. It is the early version, and it takes 95.6 seconds to run the program.

[Xerxes]

Oh, and I have a Durabrand 828 graphing calculator. It is the early version, and it takes 95.6 seconds to run the program.

Do you know, if there is a speed difference between the old HW828 or new HW1376 version?

[DrCoyote]

I don't. I have been curious to find out, but I don't have the newer (HW1376) version yet.

[Xerxes]

I've got the result for the n-queens bench of the HW1376 version already from another forum,
tested with 92 seconds. If you want to test your calculator, you will find the Durabrand 828
version in the benchmark too.

[ha1313424234]

i've just tested my two calculator:
fx-50f : 240 sec
fx-3800p: 223 sec
fx-3800p is older than fx-50f but it is a bit faster than 50f
last time i replaced the battery for fx-50f and find out it didn't have a pcb. there's just a very small ic -- silver colour, around 0.5mm^2 in size -- right below the solar panel. it has rubberized keys. fx-50f only provides 29 programming steps, so i think that's why it doesn't have a pcb. that tiny IC is the only electronic component of it! The manufacturing cost must be very low!
ooh, my calculators are the slowest!

[ha1313424234]

there is an interesting phenomenon of my calculators: fx-3800p perform x^y or x^(1/y) a bit slower than fx-50f, but it performs trigo function sin, cos , tan two times faster fx-50f. does it mean calculator also has 'weakpoint' and 'speciality' on some specific function?

[Xerxes]

I would say yes in general, because it also depends on the algorithm used to compute
the mathematical functions.

If you are interested in a speed test independently from the mathematical functions, try this:

-  355  FX-180P	  Kout 1 + 1 = Kin 1 Kout 1 x>0
 -  360  FX-3600P	 Kout 1 + 1 = Kin 1 Kout 1 x>0
 -  825  FX-3650P	 Lbl 0 A+1->A A>0=>Goto 0
 - 2810  FX-3900Pv	Kout 1 + 1 = Kin 1 Kout 1 x>0

Store zero in K1 or A and start the program. Interrupt the execution exactly after 1 minute
and recall K1 or A. I guess the speed of FX-50F is similar to the FX-180P or FX-3600P, but
I'm not sure about the FX-3800P.

[McCoy]

Acoording to this comparison shown on that link, does it mean that ti93 is faster than both ti89 and classpad?

[DrCoyote]

I have checked speed of some of programmable calculators on the time expensive power function.

My HP 35s runs it in about 47 seconds in RPN mode. This seems odd, as you get 27.8 seconds on the 33s. I'll have to check it out on my 33s, as I find it hard to believe that the 33s is that much faster.

Could you please share the code you used on the 33s, Sergei?

[Xerxes]

@McCoy:

The speed of the TI calculators with 68000 inside (TI-89/92/V200) is almost the same.
The HW1 models are a bit slower and the HW3 models are usually slightly faster compared
to the widespeaded HW2 models.

The reason for the difference of the TI-92 and the TI-89 in Sergei's test maybe caused
by different mode(Auto/Exact/Approx) settings.



@DrCoyote:

I have made some tests on the 33S and 35S and I can confirm the higher speed of the 33S.
For example in the n-queens test the 33S needs 2:11 but the 35S 4:17 although I have already
optimized the 35S code by avoiding constants in loops because these are very slow on the 35S.

The Hardware of both calculators is the same (6502 core @ 4-5 MHz), but the 35S has a much
more complex system and a new but slower keystroke programming language.

The HW828 version of the Durabrand 828 is a clon of the Citizen SRP-320G.

[DrCoyote]

I borrowed a Datexx DS-883 from a friend. It's a pleasant little machine, if a bit wierd. Anyway, Sergei, it runs your benchmark in 51.2 seconds.

...And the TRS-80 Pocket Computer 2 (Sharp PC-1500) runs the benchmark in 59.6 seconds.

A Radio Shack Pocket Computer 1 (Sharp PC 1211) takes 272 seconds.

A Sharp EL-9200C does it in 27.1 seconds.

The Corner Office ATC-139 runs the benchmark in 29.8 seconds.

[Leszek]

If you are interested in a speed test independently from the mathematical functions, try this:

-  355  FX-180P	  Kout 1 + 1 = Kin 1 Kout 1 x>0
-  360  FX-3600P	 Kout 1 + 1 = Kin 1 Kout 1 x>0
-  825  FX-3650P	 Lbl 0 A+1->A A>0=>Goto 0
- 2810  FX-3900Pv	Kout 1 + 1 = Kin 1 Kout 1 x>0

Store zero in K1 or A and start the program. Interrupt the execution exactly after 1 minute
and recall K1 or A.

I've tested it on my CASIO fx-4800P.
The result is: 3969

Leszek

[Leszek]

</SPAN></FONT>

Hi Siergiej,

Thanks a lot for such a big speed
comparision. It is very useful.

I have a CASIO fx-4800P and I didn't notice
it in your table. It's my code:

===========================================================
I=10:Lbl 1:A=1.0000001:B=A:J=27:Lbl
2:A=A´A:B=B^2.01:Dsz J:Goto 2:Dsz I:Goto
1
A
B
===========================================================

And time is: 17,2 s

It's interesting, because it's much better
time than newer and richer CASIO fx-5800P.

Answers:
A = 674 529,1097
B = 4 669 408,348

[Xerxes]

Hi Leszek,

thanks for testing, but the FX-4800P is already present in the n-queens benchmark. I have used the simple but less accurate test above only in case of lack of indirect addressing.
The FX-4800P and FX-4850P are also faster than the FX-5800P using this test.

[bangonkali]

Mcl
 Fixm
 R=8  (8 queens)
 Lbl 0
 X=R=>Goto 4_
 X=X+1
 A[X]=R
 Lbl 1
 S=S+1
 Y=X
 Lbl 2
 Y=Y-1
 Y=0=>Goto 0_
 T=A[X]-A[Y]
 T=0=>Goto 3_
 X-Y<>Abs T=>Goto 2_
 Lbl 3
 A[X]=A[X]-1
 Ans<>0=>Goto 1_
 X=X-1
 X<>0=>Goto 3_
 Lbl 4
 S

hi. im a newbie here. i have Casio 4500PA, i tried the code, but it doesn't seem to work. when i run the program, the output is 8, and it comes out directly, just 8.

[Xerxes]

Hi Bangonkali,

The language of the FX-4500P(A) is very close to the FX-4800P/4850P but
there is no DSZ command.

Probably the test code doesn't work, because you haven't used the end sign
at the comparisons ( 2ndF + SQRT ). I've used the underscore in the listing
but its represented as a white triangle in the display.